Research map / Video / Video generation
LiveEdit: Towards Real-Time Diffusion-Based Streaming Video Editing
Xinyu Wang, Chongbo Zhao, Fangneng Zhan, Yue Ma
Read the paperCheck your idea against 137,217 papers
AI-extracted. The measured effects below were read by AI from the paper's full text (no summary yet). Numbers and claims can be wrong or missing context, so check the paper before you cite or build on it.
Measured effects reported in the paper
| What was changed | Where | Measured | Effect |
|---|---|---|---|
| add Stage 3 DMD step distillation to teacher-forced causal model (ablation) | LiveEdit · Distillation stage ablation | Latency: 200.36 -> 7.89 | Improved |
| add Stage 3 DMD distillation (ablation) | LiveEdit · Distillation stage ablation | Image Quality: 0.702 -> 0.720 | Improved |
| add Stage 3 DMD distillation (ablation) | LiveEdit · Distillation stage ablation | Text Alignment: 0.264 -> 0.265 | No clear effect |
| add chunk-wise causal teacher forcing to bidirectional foundation model (ablation) | LiveEdit · Distillation stage ablation | Text Alignment: 0.268 -> 0.264 | Hurt |
| add chunk-wise causal teacher forcing (ablation) | LiveEdit · Distillation stage ablation | Image Quality: 0.716 -> 0.702 | Hurt |
| add token cache on self-attention layers (ablation) | LiveEdit · Streaming video editing benchmark (120 pairs) | TA: 0.265 -> 0.270 | Improved |
| add token cache on self-attention layers (ablation) | LiveEdit · Streaming video editing benchmark (120 pairs) | BC: 0.956 -> 0.956 | No clear effect |
| add token cache on self-attention layers (ablation) | LiveEdit · Streaming video editing benchmark (120 pairs) | DD: 0.282 -> 0.256 | Hurt |
| add token cache on self-attention layers (ablation) | LiveEdit · Streaming video editing benchmark (120 pairs) | IQ: 0.720 -> 0.708 | Hurt |
| apply cache to FFN instead of self-attention (ablation) | LiveEdit · Cache placement ablation | TA: 0.270 -> 0.236 | Hurt |
| apply cache to FFN instead of self-attention (ablation) | LiveEdit · Cache placement ablation | BC: 0.956 -> 0.841 | Hurt |
| apply cache to FFN instead of self-attention (ablation) | LiveEdit · Cache placement ablation | IQ: 0.708 -> 0.513 | Hurt |
Earlier methods this paper compares with
Named baselines from earlier accepted papers in this paper's results tables: better than 1 of 1 on most shared results.
- VideoCoF — VideoCoF: Unified Video Editing with Temporal Reasoner (CVPR 2026) · better on 3 of 3 shared results
Matched by method name in results tables read by AI. Authors' own tables tend to favour the newer method, and a comparison can be missed when a method is renamed.
Closest papers by meaning
- AR-Edit: Training-Free Streaming Video Editing without Inversion (NeurIPS 2026 · similarity 0.91)
- StreamGVE: Training-Free Video Editing via Few-Step Streaming Video Generation (ECCV 2026 · similarity 0.89)
- ProEdit: Simple Progression is All You Need for High-Quality 3D Scene Editing (NeurIPS 2024 · similarity 0.86)
- AdaState: Self-Evolving Anchors for Streaming Video Generation (NeurIPS 2026 · similarity 0.85)
- CCEdit: Creative and Controllable Video Editing via Diffusion Models (CVPR 2024 · similarity 0.85)
- VINCIE-NExT: Unlocking Video Editing from Images via In-Context Modeling (NeurIPS 2026 · similarity 0.85)
Writing a paper in this area? scime draft is a free LaTeX editor with AI included, real-time co-authors and a check of your references and venue rules before you submit. It also imports OpenAI Prism projects.
